Supervised Learning Occurs in Visual Perceptual Learning of Complex Natural Images

Perceptual learning improves discrimination but does not reduce distortions in appearance

Human perceptual sensitivity often improves with training, a phenomenon known as "perceptual learning." Another important perceptual dimension is appearance, the subjective sense of stimulus magnitude. Are training-induced improvements in sensitivity accompanied by more accurate appearance? Here, we examined this question by measuring both discrimination (sensitivity) and estimation (appearance) responses to near-horizontal motion directions, which are known to be repulsed away from horizontal. Participants performed discrimination and estimation tasks before and after training in either the discrimination or the estimation task or none (control group). Human observers who trained in either discrimination or estimation exhibited improvements in discrimination accuracy, but estimation repulsion did not decrease; instead, it either persisted or increased. Hence, distortions in perception can be exacerbated after perceptual learning. We developed a computational observer model in which perceptual learning arises from increases in the precision of underlying neural representations, which explains this counterintuitive finding. For each observer, the fitted model accounted for discrimination performance, the distribution of estimates, and their changes with training. Our empirical findings and modeling suggest that learning enhances distinctions between categories, a potentially important aspect of real-world perception and perceptual learning.

A neural geometry approach comprehensively explains apparently conflicting models of visual perceptual learning

Visual perceptual learning (VPL), defined as long-term improvement in a visual task, is considered a crucial tool for elucidating underlying visual and brain plasticity. Previous studies have proposed several neural models of VPL, including changes in neural tuning or in noise correlations. Here, to adjudicate different models, we propose that all neural changes at single units can be conceptualized as geometric transformations of population response manifolds in a high-dimensional neural space. Following this neural geometry approach, we identified neural manifold shrinkage due to reduced trial-by-trial population response variability, rather than tuning or correlation changes, as the primary mechanism of VPL. Furthermore, manifold shrinkage successfully explains VPL effects across artificial neural responses in deep neural networks, multivariate blood-oxygenation-level-dependent signals in humans and multiunit activities in monkeys. These converging results suggest that our neural geometry approach comprehensively explains a wide range of empirical results and reconciles previously conflicting models of VPL.

Artificial intelligence in pathologic diagnosis, prognosis and prediction of prostate cancer

Histopathology, which is the gold-standard for prostate cancer diagnosis, faces significant challenges. With prostate cancer ranking among the most common cancers in the United States and worldwide, pathologists experience an increased number for prostate biopsies. At the same time, precise pathological assessment and classification are necessary for risk stratification and treatment decisions in prostate cancer care, adding to the challenge to pathologists. Recent advancement in digital pathology makes artificial intelligence and learning tools adopted in histopathology feasible. In this review, we introduce the concept of AI and its various techniques in the field of histopathology. We summarize the clinical applications of AI pathology for prostate cancer, including pathological diagnosis, grading, prognosis evaluation, and treatment options. We also discuss how AI applications can be integrated into the routine pathology workflow. With these rapid advancements, it is evident that AI applications in prostate cancer go beyond the initial goal of being tools for diagnosis and grading. Instead, pathologists can provide additional information to improve long-term patient outcomes by assessing detailed histopathologic features at pixel level using digital pathology and AI. Our review not only provides a comprehensive summary of the existing research but also offers insights for future advancements.

A hole in a piece of cardboard and predictive brain: the incomprehension of modern art in the light of the predictive coding paradigm

Incomprehension of and resistance to contemporaneous art have been constant features in the development of modern art. The predictive coding framework can be used to analyse this response by outlining the difference between the misunderstanding of (i) contemporary conceptual/minimalist art and (ii) early modern avant-garde art and by elucidating their underlying cognitive mechanisms. In both of these cases, incomprehension and its behavioural consequences are tied to the failure of the optimal prediction error (PE) minimization that is involved in the perception of such art works. In the case of contemporary conceptual/minimalist art the failure stems from the fact that the encounter results in non-salient visual sensations and generates no PE. In early modern avant-garde art, the occasional inability of viewers to recognize pictorial content using new pictorial conventions reflected the absence of suitable priors to explain away ambiguous sensory data. The capacity to recognize pictorial content in modernist painting, as a prerequisite for a satisfying encounter with such works and ultimately a wider acceptance of new artistic styles, required an updating of a number of expectations in order to optimize the fit between priors and sensations, from low-level perceptual priors to the development of higher-level, culturally determined expectations. This article is part of the theme issue 'Art, aesthetics and predictive processing: theoretical and empirical perspectives'.

Enhancing visual perceptual learning using transcranial electrical stimulation: Transcranial alternating current stimulation outperforms both transcranial direct current and random noise stimulation

Diverse strategies can be employed to enhance visual skills, including visual perceptual learning (VPL) and transcranial electrical stimulation (tES). Combining VPL and tES is a popular method that holds promise for producing significant improvements in visual acuity within a short time frame. However, there is still a lack of comprehensive evaluation regarding the effects of combining different types of tES and VPL on enhancing visual function, especially with a larger sample size. In the present study, we recruited four groups of subjects (26 subjects each) to learn an orientation discrimination task with five daily training sessions. During training, the occipital region of each subject was stimulated by one type of tES-anodal transcranial direct current stimulation (tDCS), alternating current stimulation (tACS) at 10 Hz, high-frequency random noise stimulation (tRNS), and sham tACS-while the subject performed the training task. We found that, compared with the sham stimulation, both the high-frequency tRNS and the 10-Hz tACS facilitated VPL efficiently in terms of learning rate and performance improvement, but there was little modulatory effect in the anodal tDCS condition. Remarkably, the 10-Hz tACS condition exhibited superior modulatory effects compared with the tRNS condition, demonstrating the strongest modulation among the most commonly used tES types for further enhancing vision when combined with VPL. Our results suggest that alpha oscillations play a vital role in VPL. Our study provides a practical guide for vision rehabilitation.

Informational feedback accelerates learning in multi-alternative perceptual judgements of orientation

Experience or training can substantially improve perceptual performance through perceptual learning, and the extent and rate of these improvements may be affected by feedback. In this paper, we first developed a neural network model based on the integrated reweighting theory (Dosher et al., 2013) to account for perceptual learning and performance in n-alternative identification tasks and the dependence of learning on different forms of feedback. We then report an experiment comparing the effectiveness of response feedback (RF) versus accuracy feedback (AF) or no feedback (NF) (full versus partial versus no supervision) in learning a challenging eight-alternative visual orientation identification (8AFC) task. Although learning sometimes occurred in the absence of feedback (NF), RF had a clear advantage above AF or NF in this task. Using hybrid supervision learning rules, a new n-alternative identification integrated reweighting theory (I-IRT) explained both the differences in learning curves given different feedback and the dynamic changes in identification confusion data. This study shows that training with more informational feedback (RF) is more effective, though not necessary, in these challenging n-alternative tasks, a result that has implications for developing training paradigms in realistic tasks.

Artificial intelligence-based radiomics in bone tumors: Technical advances and clinical application

Radiomics is the extraction of predefined mathematic features from medical images for predicting variables of clinical interest. Recent research has demonstrated that radiomics can be processed by artificial intelligence algorithms to reveal complex patterns and trends for diagnosis, and prediction of prognosis and response to treatment modalities in various types of cancer. Artificial intelligence tools can utilize radiological images to solve next-generation issues in clinical decision making. Bone tumors can be classified as primary and secondary (metastatic) tumors. Osteosarcoma, Ewing sarcoma, and chondrosarcoma are the dominating primary tumors of bone. The development of bone tumor model systems and relevant research, and the assessment of novel treatment methods are ongoing to improve clinical outcomes, notably for patients with metastases. Artificial intelligence and radiomics have been utilized in almost full spectrum of clinical care of bone tumors. Radiomics models have achieved excellent performance in the diagnosis and grading of bone tumors. Furthermore, the models enable to predict overall survival, metastases, and recurrence. Radiomics features have exhibited promise in assisting therapeutic planning and evaluation, especially neoadjuvant chemotherapy. This review provides an overview of the evolution and opportunities for artificial intelligence in imaging, with a focus on hand-crafted features and deep learning-based radiomics approaches. We summarize the current application of artificial intelligence-based radiomics both in primary and metastatic bone tumors, and discuss the limitations and future opportunities of artificial intelligence-based radiomics in this field. In the era of personalized medicine, our in-depth understanding of emerging artificial intelligence-based radiomics approaches will bring innovative solutions to bone tumors and achieve clinical application.

Protocol to conduct functional magnetic resonance spectroscopy in different age groups of human participants

We present a protocol to conduct functional magnetic resonance spectroscopy (fMRS) in human participants before, during, and after training on a visual task. We describe steps for participant setup, volume-of-interest placement, fMRS measurement, and post-scan tests. We discuss the design, analysis, and interpretation of fMRS experiments. This protocol can be adapted to investigate the dynamics of chief excitatory and inhibitory neurotransmitters (glutamate and γ-aminobutyric acid, GABA, respectively) while participants perform or learn perceptual, motor, or cognitive tasks. For complete details on the use and execution of this protocol, please refer to Frank et al. (2022).1.

Perceptual learning: Training together makes us better

Perceptual learning is often thought to rely primarily on low-level visual areas. A new study shows that learning with a partner can improve perceptual learning performance, demonstrating that higher cognitive processes play a larger role in perceptual learning than previously supposed.

Evaluating the effectiveness of different perceptual training methods in a difficult visual discrimination task with ultrasound images

Recent work has shown that perceptual training can be used to improve the performance of novices in real-world visual classification tasks with medical images, but it is unclear which perceptual training methods are the most effective, especially for difficult medical image discrimination tasks. We investigated several different perceptual training methods with medically naïve participants in a difficult radiology task: identifying the degree of hepatic steatosis (fatty infiltration of the liver) in liver ultrasound images. In Experiment 1a (N = 90), participants completed four sessions of standard perceptual training, and participants in Experiment 1b (N = 71) completed four sessions of comparison training. There was a significant post-training improvement for both types of training, although performance was better when the trained task aligned with the task participants were tested on. In both experiments, performance initially improves rapidly, with learning becoming more gradual after the first training session. In Experiment 2 (N = 200), we explored the hypothesis that performance could be improved by combining perceptual training with explicit annotated feedback presented in a stepwise fashion. Although participants improved in all training conditions, performance was similar regardless of whether participants were given annotations, or underwent training in a stepwise fashion, both, or neither. Overall, we found that perceptual training can rapidly improve performance on a difficult radiology task, albeit not to a comparable level as expert performance, and that similar levels of performance were achieved across the perceptual training paradigms we compared.

Machine learning approach identified clusters for patients with low cardiac output syndrome and outcomes after cardiac surgery

BACKGROUND

Low cardiac output syndrome (LCOS) is the most serious physiological abnormality with high mortality for patients after cardiac surgery. This study aimed to explore the multidimensional data of clinical features and outcomes to provide individualized care for patients with LCOS.

METHODS

The electronic medical information of the intensive care units (ICUs) was extracted from a tertiary hospital in South China. We included patients who were diagnosed with LCOS in the ICU database. We used the consensus clustering approach based on patient characteristics, laboratory data, and vital signs to identify LCOS subgroups. The consensus clustering method involves subsampling from a set of items, such as microarrays, and determines to cluster of specified cluster counts (k). The primary clinical outcome was in-hospital mortality and was compared between the clusters.

RESULTS

A total of 1,205 patients were included and divided into three clusters. Cluster 1 (n = 443) was defined as the low-risk group [in-hospital mortality =10.1%, odds ratio (OR) = 1]. Cluster 2 (n = 396) was defined as the medium-risk group [in-hospital mortality =25.0%, OR = 2.96 (95% CI = 1.97-4.46)]. Cluster 3 (n = 366) was defined as the high-risk group [in-hospital mortality =39.2%, OR = 5.75 (95% CI = 3.9-8.5)].

CONCLUSION

Patients with LCOS after cardiac surgery could be divided into three clusters and had different outcomes.

Transfer of Tactile Learning from Trained to Untrained Body Parts Supported by Cortical Coactivation in Primary Somatosensory Cortex

A pioneering study by Volkmann (1858) revealed that training on a tactile discrimination task improved task performance, indicative of tactile learning, and that such tactile learning transferred from trained to untrained body parts. However, the neural mechanisms underlying tactile learning and transfer of tactile learning have remained unclear. We trained groups of human subjects (female and male) in daily sessions on a tactile discrimination task either by stimulating the palm of the right hand or the sole of the right foot. Task performance before training was similar between the palm and sole. Posttraining transfer of tactile learning was greater from the trained right sole to the untrained right palm than from the trained right palm to the untrained right sole. Functional magnetic resonance imaging (fMRI) and multivariate pattern classification analysis revealed that the somatotopic representation of the right palm in contralateral primary somatosensory cortex (SI) was coactivated during tactile stimulation of the right sole. More pronounced coactivation in the cortical representation of the right palm was associated with lower tactile performance for tactile stimulation of the right sole and more pronounced subsequent transfer of tactile learning from the trained right sole to the untrained right palm. In contrast, coactivation of the cortical sole representation during tactile stimulation of the palm was less pronounced and no association with tactile performance and subsequent transfer of tactile learning was found. These results indicate that tactile learning may transfer to untrained body parts that are coactivated to support tactile learning with the trained body part.SIGNIFICANCE STATEMENT Perceptual skills such as the discrimination of tactile cues can improve by means of training, indicative of perceptual learning and sensory plasticity. However, it has remained unclear whether and if so, how such perceptual learning can occur if the training task is very difficult. Here, we show for tactile perceptual learning that the representation of the palm of the hand in primary somatosensory cortex (SI) is coactivated to support learning of a difficult tactile discrimination task with tactile stimulation of the sole of the foot. Such cortical coactivation of an untrained body part to support tactile learning with a trained body part might be critically involved in the subsequent transfer of tactile learning between the trained and untrained body parts.

Simple contextual cueing prevents retroactive interference in short-term perceptual training of orientation detection tasks

Perceptual training of multiple tasks suffers from interference between the trained tasks. Here, we conducted five psychophysical experiments with separate groups of participants to investigate the possibility of preventing the interference in short-term perceptual training. We trained the participants to detect two orientations of Gabor stimuli in two adjacent days at the same retinal location and examined the interference of training effects between the two orientations. The results showed significant retroactive interference from the second orientation to the first orientation (Experiment 1 and Experiment 2). Introducing a 6-h interval between the pre-test and training of the second orientation did not eliminate the interference effect, excluding the interpretation of disrupted reconsolidation as the pre-test of the second orientation may reactivate and destabilize the representation of the first orientation (Experiment 3). Finally, the training of the two orientations was accompanied by fixations in two colors, each serving as a contextual cue for one orientation. The results showed that the retroactive interference was not evident if the participants passively perceived contextual cues during the training and test sessions (Experiment 4). Importantly, this facilitation effect could be observed if the contextual cues appeared only during the training, demonstrating the robustness of the effect (Experiment 5). Our findings suggest that the retroactive interference effect in short-term perceptual training of orientation detection tasks was likely the result of higher-level factors such as shared contextual cues embedded in the tasks. The efficiency of multiple perceptual trainings could be facilitated by associating the trained tasks with different contextual cues.

Identifying Long- and Short-Term Processes in Perceptual Learning

Practice makes perfect in almost all perceptual tasks, but how perceptual improvements accumulate remains unknown. Here, we developed a multicomponent theoretical framework to model contributions of both long- and short-term processes in perceptual learning. Applications of the framework to the block-by-block learning curves of 49 adult participants in seven perceptual tasks identified ubiquitous long-term general learning and within-session relearning in most tasks. More importantly, we also found between-session forgetting in the vernier-offset discrimination, face-view discrimination, and auditory-frequency discrimination tasks; between-session off-line gain in the visual shape search task; and within-session adaptation and both between-session forgetting and off-line gain in the contrast detection task. The main results of the vernier-offset discrimination and visual shape search tasks were replicated in a new experiment. The multicomponent model provides a theoretical framework to identify component processes in perceptual learning and a potential tool to optimize learning in normal and clinical populations.

A protocol to examine the learning effects of 'multisystem mapping' training combined with post-training sleep consolidation in beginning readers

We have recently used randomized controlled trials to examine the impact of a short neuroscience-informed causal intervention using a targeted training to inhibit a deeply rooted visual mechanism (mirror invariance) that hinders literacy acquisition, combined with post-training sleep (for learning consolidation). Using this training protocol, we have shown unprecedented improvements in visual perception of letters, writing, and a two-fold increase in reading fluency in first graders. Here, we describe this ecologically valid school-based intervention protocol to probe inhibition of mirror invariance for letters, including the detailed training instructions, post-training sleep consolidation, as well as practical tips and potential adaptations to different school sizes. For complete details on the use and execution of this protocol, please refer to Torres et al., (2021).

A behavioral training protocol using visual perceptual learning to improve a visual skill

We describe a behavioral training protocol using visual perceptual learning (VPL) to improve visual detection skills in non-experts for subtle mammographic lesions indicative of breast cancer. This protocol can be adapted for the professional training of experts (radiologists) or to improve visual skills for other tasks, such as the detection of targets in photo or video surveillance.

For complete details on the use and execution of this protocol, please refer to Frank et al. (2020a).

•

Behavioral training using VPL induces long-lasting improvements of a visual skill

•

Training should be conducted using detailed feedback about response accuracy

•

Training can be conducted with minimal technical equipment

•

VPL protocol can be used for clinical or other professional training

Fundamental Differences in Visual Perceptual Learning between Children and Adults

It has remained uncertain whether the mechanisms of visual perceptual learning (VPL)^1-4 remain stable across the lifespan or undergo developmental changes. This uncertainty largely originates from missing results about the mechanisms of VPL in healthy children. We here investigated the mechanisms of task-irrelevant VPL in healthy elementary school age children (7-10 years old) and compared their results to healthy young adults (18-31 years old). Subjects performed a rapid-serial-visual-presentation (RSVP) task at central fixation over the course of several daily sessions while coherent motion was merely exposed as a task-irrelevant feature in the visual periphery either at threshold or suprathreshold levels for coherent motion detection. As a result of this repeated exposure, children and adults both showed enhanced discrimination performance for the threshold task-irrelevant feature as in previous studies with adults.^5-8 However, adults demonstrated a decreased performance for the suprathreshold task-irrelevant feature whereas children increased performance. One possible explanation for this difference is that children cannot effectively suppress salient task-irrelevant features because of weaker selective attention ability compared to that of adults.^9-11 However, our results revealed to the contrary that children with stronger selective attention ability, as measured by the useful field of view (UFOV) test, showed greater increases in performance for the suprathreshold task-irrelevant feature. Together, these results suggest that the mechanisms of VPL change dramatically from childhood to adulthood due to a change in the way learners handle salient task-irrelevant features.

What do radiologists look for? Advances and limitations of perceptual learning in radiologic search

Supported by guidance from training during residency programs, radiologists learn clinically relevant visual features by viewing thousands of medical images. Yet the precise visual features that expert radiologists use in their clinical practice remain unknown. Identifying such features would allow the development of perceptual learning training methods targeted to the optimization of radiology training and the reduction of medical error. Here we review attempts to bridge current gaps in understanding with a focus on computational saliency models that characterize and predict gaze behavior in radiologists. There have been great strides toward the accurate prediction of relevant medical information within images, thereby facilitating the development of novel computer-aided detection and diagnostic tools. In some cases, computational models have achieved equivalent sensitivity to that of radiologists, suggesting that we may be close to identifying the underlying visual representations that radiologists use. However, because the relevant bottom-up features vary across task context and imaging modalities, it will also be necessary to identify relevant top-down factors before perceptual expertise in radiology can be fully understood. Progress along these dimensions will improve the tools available for educating new generations of radiologists, and aid in the detection of medically relevant information, ultimately improving patient health.

Early Visual Cortex Stimulation Modifies Well-Consolidated Perceptual Gains

Perception thresholds can improve through repeated practice with visual tasks. Can an already acquired and well-consolidated perceptual skill be noninvasively neuromodulated, unfolding the neural mechanisms involved? Here, leveraging the susceptibility of reactivated memories ranging from synaptic to systems levels across learning and memory domains and animal models, we used noninvasive brain stimulation to neuromodulate well-consolidated reactivated visual perceptual learning and reveal the underlying neural mechanisms. Subjects first encoded and consolidated the visual skill memory by performing daily practice sessions with the task. On a separate day, the consolidated visual memory was briefly reactivated, followed by low-frequency, inhibitory 1 Hz repetitive transcranial magnetic stimulation over early visual cortex, which was individually localized using functional magnetic resonance imaging. Poststimulation perceptual thresholds were measured on the final session. The results show modulation of perceptual thresholds following early visual cortex stimulation, relative to control stimulation. Consistently, resting state functional connectivity between trained and untrained parts of early visual cortex prior to training predicted the magnitude of perceptual threshold modulation. Together, these results indicate that even previously consolidated human perceptual memories are susceptible to neuromodulation, involving early visual cortical processing. Moreover, the opportunity to noninvasively neuromodulate reactivated perceptual learning may have important clinical implications.

Perceptual Expertise: How Is It Achieved?

Humans are perceptual experts and we are constantly refining how we detect and discriminate objects in the world around us, often without any explicit instruction. But instruction can be helpful and sometimes even necessary. New research highlights the importance of instruction for achieving robust long-term retention of learning to identify complex features in natural images such as those used in radiology.